plan-improve-apply

dp:plan-improve-apply

Apply review fixes to plan.md. Self-contained: embeds the original /plan-improve-apply discipline verbatim, then extends with dp-specific rules. Does NOT invoke /plan-improve-apply.

Inputs

• RUN_DIR — run directory.
• <RUN_DIR>/review.md — issues from dp:plan-improve.
• <RUN_DIR>/plan.md — target file to patch.
• <RUN_DIR>/state.json — for autonomous flag.

If review.md says No issues found., skip to step 5 (advance) and hand off.

Procedure

Session id: if a DP_SESSION_ID=<id> line is present in your conversation context (see the orchestrator command's "Session id capture and propagation" section for the matching rule), substitute that value for every <DP_SESSION_ID> placeholder in the bun commands below. If the line is not in context, drop the --session "<DP_SESSION_ID>" argument entirely; advance.ts falls back to process.env.DP_SESSION_ID.

1. Mark running

bun ${DP_PLUGIN_ROOT}/scripts/cli/advance.ts set <RUN_DIR> steps.plan-improve-apply.status running --session "<DP_SESSION_ID>"

2. Original /plan-improve-apply discipline (verbatim — follow exactly)

Create a todo for each issue found in review, and one final todo: "Check plan consistency". Work on each todo one by one, think about the issue, ask interactive question(s) with proper explanation if needed, update the plan, then proceed to next todo.

For final "Check plan consistency", read the full plan and understand if small plan patches that were done correlate with each other and no conflicts.

No questions, decide all yourself if initial prompt contains "auto".

For dp:

• Read issues from <RUN_DIR>/review.md.
• Use TodoWrite to create one todo per issue + a final "Check plan consistency" todo.
• Work each todo one by one, patching <RUN_DIR>/plan.md in place.
• "auto" mode is determined by state.json.autonomous — read it once and apply throughout. If autonomous === true, decide everything yourself; never ask.

3. dp-specific extension — options instead of open prompts

When a fix direction for an issue is unclear and autonomous === false:

• Use AskUserQuestion with concrete options (not an open-ended prompt). Always include an explicit "Decide for me" option as the last choice — it falls back to your best-guess fix and proceeds.
• The original "ask interactive question(s) with proper explanation if needed" rule still applies, but always in the structured-options form. We want the user to point at A or B, not to type an essay.

When a fix direction is clear (the review tells you exactly what to do, or autonomous mode), just patch and move on.

4. Final todo — consistency check

Re-read the entire plan.md. Look for contradictions between sections that may have appeared after multiple patches. Fix any you find. Update file in place.

5. Mark done and advance

bun ${DP_PLUGIN_ROOT}/scripts/cli/advance.ts advance <RUN_DIR> plan-improve-apply --session "<DP_SESSION_ID>"

6. Hand off to dp:plan-wrapup — do not text-stop

The plugin's Stop hook gates progression on Claude Code (hard block while steps.plan-wrapup.status === "pending") and auto-prompts the next skill on Cursor (soft auto-submit). Either way, advancing state.json correctly is mandatory.

Print a one-liner first, referencing the patched plan.md as a markdown link:

Review fixes applied — open [plan.md](${DP_STATE_DIR}/feature-pipeline/<feature>/plan.md). Finalizing now.

On Claude Code: your very next action MUST be a Skill-tool invocation in this same turn:

Skill(skill_name = "dp:plan-wrapup")

On Cursor: there is no Skill tool — end your turn after the one-liner above. The plugin's stop hook will auto-submit /plan-wrapup as a follow-up turn, triggering the next skill via slash-prefix auto-discovery.